Alleles at different loci are not always inherited independently — those in linkage disequilibrium (LD) occur together in populations more often than is expected. The extent to which LD occurs in the human genome, and how this affects variation, will determine how easy it will be to map complex-trait loci through whole-genome association studies. Now, Stumpf and Goldstein use computer simulations to examine the block-like structure of LD and conclude that it is time to abandon the general idea of an average extent of LD in the human genome.

Initial assessments of how to use LD in genetic-association studies assumed a uniform recombination rate and an idealized demographic population history. But, theory is seldom exactly like real life. There has been much evidence that recombination rates vary across the genome, sometimes resulting in a block-like genome structure. Stumpf and Goldstein set out to model the effects of recombination hotspots, as well as demography, on the extent of the LD. To assess the interactions between recombination hotspots and demography the authors used simulations of populations that undergo several bottle-necks that create LD and observed how the associations decayed under different intensities of recombination hotspots. The results show that the probability of the LD block-like structure is intimately linked with demography — severe bottlenecks delay the block-like genome structure whereas relatively high intensities of hotspots maintain it for long periods of time.

Not satisfied with simulations alone, Stumpf and Goldstein turned to real data. Having considered models of populations with different demographic histories — Europeans, Finns and Georgian Jews — the authors conclude that as a result of a strong interaction between demography and hotspots, the block-like structure of LD might be present in some populations but absent from others. On this basis, they suggest that different statistical methods will be needed for association studies in different populations and/or gene regions. They also suggest that, depending on their demographies, different populations might be suitable for different stages of association mapping. The LD debate is far from over, but it is clear that we must be cautious about extrapolating from one gene region, and one population, to another.